Bad Reasons

• It’s neat
  
• I want to show off
  
• One of my dimensions is time

Good Reasons

• I have four or more meaningful data dimensions
  
• I want to display many levels of a single dimension

Also: If you need to see all the data points at the same time DON’T use animation

• What information do you want the audience to take away?
  
• Movement leads the viewer to interpretation of the data - this can be good or bad
  
• Speed creates meaning, be conscious of that
  
• Each static frame should be informative alone as well

(Be open to the idea that you might not need animation to be most effective)

library(ggplot2) # make plots
library(gganimate) # animate the plots
library(gifski) # render gifs

Optional, maybe interesting packages

library(transformr) # additional smooth transformations
library(patchwork) # arrange multiple plots in a single layout
library(ggforce) # extended options for ggplot functionality

Frame

• Animations are a series of single images (frames) strung together (think of “frames per second”)

State

• The current grouping variable being shown. It may be one frame, or a single state representation can stay visible for multiple frames

Transition

• How the visual rendering moves from state to state

Enter/Exit

• How datapoints with no predecessor or follower are handled

Using data on displaced persons living in South Africa, show patterns and trends in refugee movement over twenty-plus years.

1. Where do most refugees come from?

   • These may be places with high instability or political turmoil at a given time
     
2. Over time, are the most common countries of origin about the same?
   

3. Is the number of refugees coming from the most common countries of origin stable?

Follow along with the notebook: https://github.com/skirmer/animating_dataviz/blob/master/ioslides_deck.Rmd

Data Source: UNHCR via data.world: UNHCR’s populations of concern residing in South Africa

Grab dataset of displaced persons living in South Africa

Select the top 10 countries of origin for each year, apply some filters.

library(magrittr)
library(dplyr)

plotDT <- dtset %>%
  filter(X_country_origin != "Various/Unknown") %>%
  filter(X_population_type == "Refugees (incl. refugee-like situations)") %>%
  group_by(X_date_year) %>%
  mutate(rank = rank(-X_affected, ties.method = "first") * 1) %>%
  ungroup() %>%
  filter(rank <= 10) %>%
  data.frame()

Group by country, bar height is affected persons

baseplot1 <- ggplot(plotDT, 
aes(X_date_year, 
    group = X_country_origin, 
    fill = as.factor(X_country_origin), 
    color = as.factor(X_country_origin)))+
theme_bw()+
theme(legend.position = "bottom")+
geom_bar(aes(y = X_affected), stat = "identity", position = "dodge")

Ew. This is not effective.

Let’s make this!

baseplot2 <- baseplot1 +
  coord_flip(clip = "off", expand = FALSE, ylim = c(0, 50000))

Can’t use the same stub now.

baseplot3 <- ggplot(plotDT, 
  aes(x=rank, 
    group = X_country_origin, 
    fill = as.factor(X_country_origin), 
    color = as.factor(X_country_origin)))+
theme_bw()+
theme(legend.position = "bottom")+
geom_text(aes(y = 0, label = paste(X_country_origin, " ")), vjust = 0.2, hjust = 1) +
coord_flip(clip = "off", expand = FALSE, ylim = c(0, 50000)) +
  geom_bar(aes(y = X_affected), stat = "identity", position = "identity")

Left side: grouped bar, sorted by year
Right side: not grouped bar, sorted by rank

If your stub has a theme() segment, applying a new one will overrule it.

baseplot3 <- baseplot3 + 
  theme(legend.position = "none",
    axis.ticks.y = element_blank(), 
    axis.text.y  = element_blank(), 
    axis.title.y = element_blank(),
    plot.margin = margin(1,1,1,5, "cm"))

baseplot3 <- baseplot3 +
scale_y_continuous(labels = scales::comma) +
scale_x_reverse()

Now we have the different “frames” all layered on top of each other.

Revisit our goals:

1. Where do most refugees come from?
   
2. Over time, are the most common countries of origin about the same?
   
3. Is the number of refugees coming from the most common countries of origin stable?

Let’s try static options one last time just to be sure.

No.

Definitely not.

Literally add one more line of code to your ggplot object.

animp <- baseplot3 +
  transition_states(X_date_year)

Solving Problem 1 and 2: Added a descriptive title/label that indicate the year of the frame, label bars

animp <- baseplot3 +
  geom_text(aes(y = X_affected, 
    label = as.character(X_affected)), 
    color = "black", vjust = 0.2, hjust = -.5)+
  labs(title = "Refugees Residing in South Africa by Origin, {closest_state}"
    , y="Affected Persons")+
  transition_states(X_date_year)

Solving Problem 3: how do we want the animation elements to move?

Option: shrink and grow on exit and enter

animp <- baseplot3 +
  geom_text(aes(y = X_affected, 
    label = as.character(X_affected)), 
    color = "black", vjust = 0.2, hjust = -.5)+
  labs(title = "Refugees Residing in South Africa by Origin, {closest_state}"
    , y="Affected Persons")+
  transition_states(X_date_year)+
  enter_grow() +
  exit_shrink()

It’s interesting, but probably not serving the project objectives

Solving Problem 3: how do we want the animation elements to move?

Option: Ease between positions (moving on page, not exiting or entering)

animp <- baseplot3 +
  geom_text(aes(y = X_affected, 
    label = as.character(X_affected)), 
    color = "black", vjust = 0.2, hjust = -.5)+
  labs(title = "Refugees Residing in South Africa by Origin, {closest_state}"
  , y="Affected Persons")+
  transition_states(X_date_year)+
  ease_aes('quartic-in-out')

Makes the transition speed change as it moves (functions may be cubic, quartic, etc)

Solving Problem 3: how do we want the animation elements to move?

Option: For fun, let’s try “back” to see a springier approach

animp <- baseplot3 +
  geom_text(aes(y = X_affected, 
    label = as.character(X_affected)), 
    color = "black", vjust = 0.2, hjust = -.5)+
  labs(title = "Refugees Residing in South Africa by Origin, {closest_state}"
  , y="Affected Persons")+
  transition_states(X_date_year)+
  ease_aes('back-in-out')

Feels a little cartoony- interesting, but again perhaps not what we need

Solving Problem 3: how do we want the animation elements to move?

In addition to entry, exit, and transition easing:

• Set pace for the states/transitions
  
• Ensure no smoothing between the end and restarting

animp <- baseplot3 +
  geom_text(aes(y = X_affected, 
    label = as.character(X_affected)), 
    color = "black", vjust = 0.2, hjust = -.5)+
  labs(title = "Refugees Residing in South Africa by Origin, {closest_state}"
    , y="Affected Persons")+
  transition_states(X_date_year,transition_length = 5, 
      state_length = c(rep(.25, 21), 20), wrap = FALSE)+
  ease_aes('linear')+
  enter_fade() +
  exit_fade()

Slower pace feels smoother, and doesn’t insinuate that the last frame and the first flow into each other

animate(animp, fps = 10, duration = 20, width = 800, height = 450) 

anim_save(filename = "final_race_plot2.gif")

1. Do you need animation?

Prioritize the transmission of information effectively
Make your plot serve the audience, don’t be fancy if it’s not helpful

2. Make your animation awesome

Think carefully about transitions and speed

• Figure out the in/out transitions that make sense and are not misleading/distracting
  
• Give the eye enough time to understand each state

Get feedback and test your animation on naive viewers

https://gganimate.com/

https://ggplot2.tidyverse.org/

https://ggforce.data-imaginist.com/

https://stackoverflow.com/questions/53162821/animated-sorted-bar-chart-with-bars-overtaking-each-other/53163549 (Hat tip to Jon Spring for this awesome starting point for this kind of thing!)

Even More Examples

• https://twitter.com/WeAreRLadies/status/1123209959020220419

Similar project in D3: https://observablehq.com/@johnburnmurdoch/bar-chart-race-the-most-populous-cities-in-the-world